{"type":"link","version":"1.0","title":"A Qwen3-8B agent post-trained with GRPO reaches 77% on FHIR-AgentBench, outperforming o4-mini at 50% and GLM-5 (744B parameters) at 59%","author_name":"AI Archs","author_url":"https://ai-arch.pages.dev","provider_name":"AI Archs","provider_url":"https://ai-arch.pages.dev","url":"https://ai-arch.pages.dev/n/fhir-agent-rl-exceeds-closed-model-baseline","thumbnail_url":"https://ai-arch.pages.dev/og/fhir-agent-rl-exceeds-closed-model-baseline.png","thumbnail_width":1200,"thumbnail_height":630}